M5G 1 L5
We have determined total protein, albumin, globulins, and urea in unmodified human serum using near-infrared (NIR) spectroscopy. The distinctive NIR spectrum of the individual protein fractions allowed unique spectroscopic models to be derived for each fraction using multiple linear least-squares regression (MLR). A more sophisticated partial least-squares (PLS) regression analysis was necessary to fully extract the relevant spectral information for urea.
The serum protein data sets consisted of 235 calibration samples and 85 prediction samples. The total protein calibration yielded a multiple correlation coefficient (R) of 0.986 and a standard error of calibration (SEC) of 1.7 g/L. For the validation data set, a correlation coefficient (r) of 0.985 and standard error of prediction (SEP) of 1.7 g/L were obtained. For albumin, the values were R = 0.982 and SEC = 1.2 g/L for calibration and r = 0.984 and SEP = 1.1 g/L for prediction. Calibration results for globulins were R = 0.916 and SEC = 2.1 g/L; validation results were r = 0.885 and SEP = 2.2 g/L. The sum of the separately estimated albumin and globulin fractions agreed well with the total protein estimate. For urea, the calibration model (n = 422) provided an R = 0.994 and SEC = 0.7 mmol/L and the validation results (n = 208) were r = 0.993 and SEP = 0.8 mmol/L. The within-day and between-day precision values (CVs) were respectively, 1.2 and 1.3% for total protein, 1.7 and 1.1% for albumin, 1.5 and 1.3% for globulins, and 8.3 and 5.5% for urea.